Forecasting extremes
Extreme events, such as sudden pollution spikes, financial crashes, or health crises, can have serious and unpredictable consequences. Forecasting these events is critical, but traditional models struggle when data is unstable, complex, or fast-changing.
In 2020, Dr Leonid Bogachev and his team began developing novel machine learning models capable of forecasting future extremes, utilising a combination of machine learning and innovative statistics to deliver faster, more accurate warnings in real time.
What began as a theoretical challenge in the statistical modelling of air pollution has since evolved into a far-reaching project spanning finance, environmental science, healthcare, and legal technology.
Video summary
Impact
- Knowledge discovery: found new ways to use machine-learning models and efficient software tools
- Economic impact: created a University of Leeds spin-out business
- Environmental impact: used machine-learning and efficient software to improve environmental outcomes.
Key information
- Partners and collaborators: Research and Innovation Service, Nexus
- Disciplines: mathematics, machine learning, technology
- Investigators: Dr Leonid Bogachev.
New foundations for forecasting extremes
At the heart of Dr Bogachev’s work is a fundamental innovation: a reformulation of the classical peaks-over-threshold (POT) method used to analyse rare extreme events, such as spikes in pollution levels or financial crashes.
The conventional approach assumes a stationary data environment. However, taking air pollution as an example, Dr Bogachev said: “The stationarity assumption is untenable due to a strong dependence on temporal variations of weather and traffic.
“In this situation, real-time modelling and forecasting of extremes would appear hopeless due to the high computational cost.”
Collaborating with PhD student János Gyarmati-Szabó, he developed a new regression-based POT model. Not only did it preserve threshold stability and enable real-time computation, but it proved to be six to seven times more accurate. The applications of this innovation soon proved to be more than purely academic.
Spinning out 4-Xtra Technologies
By 2021, he and his team had developed a proof-of-concept cloud-based software prototype that combined deep learning and statistical modelling. It could handle various types of incoming data, including multivariate time series, images, texts, and more, and utilised a cloud platform to scale it for industry applications.
That year, Dr Bogachev secured funding through commercialisation grants, including those from Research England, the Northern Triangle Initiative, and EPSRC Impact Acceleration Accounts (IAA).
He said: “This ‘dream-come-true’ development was possible due to my long-term proactive search, encouraged by the School of Mathematics and supported by RIS at Nexus.”
Using this crucial funding, Dr Bogachev co-founded 4-Xtra Technologies, a University of Leeds spin-out company initially targeting the financial tech sector.
It has since expanded its reach into health technology, where its forecasting tools may soon help anticipate episodes of severe hypo- or hyperglycaemia, and law, where collaboration with legal partners is developing explainable AI systems for family law settlements.
One such project, AILES (AI-assisted Legal Settlement), is supported by the AI SuperConnector scheme and aims to automate legal processes while maintaining transparency and accountability, reaching amicable resolutions, faster outcomes, and cost savings.
Bridging research, policy, and enterprise
Dr Bogachev’s work extends beyond software. He regularly advises public and private stakeholders, including the Met Office, DEFRA, and Leeds City Council.
He also supports initiatives within the University. He is a founding co-director of LIDA’s AI Horizontal Programme and a member of the Faculty of Engineering and Physical Sciences’ Capitalising on AI Opportunities discussion group, which aims to identify strengths within the Faculty to develop a forward-looking AI strategy. He has also partnered with Leeds University Business School on fintech.
His work in data science was recognised when he was awarded a prestigious Turing Fellowship by the Alan Turing Institute (2021–2023) and received a Turing grant. This was followed by a commendation from the University of Leeds’ Research Impact and Engagement Awards in 2024, where he won the prize for ‘Emerging Economic impact.’
More on the horizon
Dr Bogachev continues to contribute to grant applications with industrial partners, such as forecasting floods in Malaysia with the British Council, addressing extreme tropical rainfall with the Met Office, and working with the NHS to support the adoption of glucose forecasting tools, with potential policy implications for diabetes care.
By his own words, AILES will “take a disruptive innovation route” to deliver AI-driven tools to law firms, focusing on smaller, underserved legal practices rather than major organisations.
Dr Bogachev also continues to help shape University-wide AI efforts, including a proposed Leeds Institute for AI Research (LIFAIR). He also aims to raise the profile of industry partnerships and continue engaging both policymakers and the wider public.
With ongoing patent applications in the EU and the US, and a pipeline of ongoing projects, his work, which began as a norm-challenging approach to theoretical data, continues to target societal, health, and economic impacts on a local and national scale.
He added: “My team and I will take every opportunity to engage with the local and national governments and the Houses of Parliament to promote our work and to outreach and inform the wider public, including local schools and academies.”
About Dr Leonid Bogachev
Dr Leonid Bogachev is a Reader in Probability within the School of Mathematics at the University of Leeds, specialising in probability, statistical physics, and statistics.
This project was sponsored by the UKRI Impact Acceleration Account.
Keywords: statistics, artificial intelligence, AI